AI Algorithm with High Diagnostic Accuracy Contributes to Improving Lung Cancer Detection Rate

According to research results published in , the assistance of artificial intelligence (AI) algorithms with high diagnostic accuracy improved the performance of radiologists in detecting lung cancer on chest radiographs, and compared human interaction with AI suggestions. Increased acceptance. RadiologyJournal of the Radiological Society of North America (RSNA).

Although AI-based image diagnosis is rapidly progressing in the medical field, the factors that affect the diagnostic decisions of radiologists in AI-assisted image interpretation have not yet been elucidated.

Researchers at Seoul National University investigated how these factors affect the detection of malignant nodules in the lung during AI-assisted reading of chest radiographs.

In this retrospective study, 30 readers, including 20 chest radiologists with 5–18 years of experience and 10 radiology residents with only 2–3 years of experience, performed 120 chest X-rays without AI. A line photograph was evaluated. Of his 120 chest radiographs evaluated, 60 were from lung cancer patients (32 men) and 60 were from controls (36 men). The median age of patients was 67 years. In the second session, each group reinterpreted his X-rays with the assistance of a high-precision or low-precision AI. The reader was unaware of the fact that two different AIs were used.

By adopting high-precision AI, reader detection performance has improved significantly compared to low-precision AI. The use of high-precision AI has led to more frequent changes in reader judgment, a concept known as susceptibility.

The relatively large sample size of this study may have increased reader confidence in the AI ​​proposals. We believe that the question of human trust in AI is what we observed in the susceptibility of this study. Humans become more susceptible to the influence of AI when using AI with high diagnostic performance. “

Chang Min Park, MD, Ph.D., Lead Study Author, Department of Radiology and Institute of Radiology, Seoul National University College of Medicine

Compared to the first reading session, readers assisted by high diagnostic accuracy AI in the second reading session showed higher sensitivity (0.63 vs. 0.53) and specificity (0.94 vs. 0.88) per lesion. it was done. Alternatively, an AI-assisted reader with low diagnostic accuracy in the second reading session showed no improvement between her two reading sessions in any of these measures.

“Our research suggests that AI can help radiologists, but only if their diagnostic performance matches or exceeds that of humans,” Park said. said Dr.

This result highlights the importance of using AI with high diagnostic performance. However, Dr. Park noted that the definition of “high diagnostic performance AI” may vary depending on the task and clinical situation in which the AI ​​is used. For example, an AI model that can detect all abnormalities in a chest x-ray might seem ideal. In practice, however, such a model would be of limited value in reducing the workload in the pulmonary tuberculosis population screening setting.

“Therefore, our research requires both the development of high-performance AI models for specific tasks and consideration of the relevant clinical settings in which the AI ​​is applied, for the appropriate clinical use of AI.” ,” said Dr. Park.

In the future, the researchers hope to extend their work on human-AI collaboration to other abnormalities in chest x-rays and CT images.